AI is changing robotics. Teach robots through demonstrations, not code.
Foundation models can learn manipulation tasks from human demonstrations. New models ship every month, each better than the last. But trying them on real hardware means wrestling with vendor-specific APIs, incompatible data formats, and integration work that starts from scratch every time.
Positronic lets you use the best AI models available – on your hardware, for your tasks. One open-source Python library. Switch models without switching stacks.
How it works
- Collect – teleoperate in simulation or on hardware (phone, VR, leader arm)
- Train – fine-tune on your data. Switch models without re-recording.
- Run – unified inference across vendors. Same code, any model, any robot.
- Iterate – monitor, collect edge cases, retrain.
Store your data once, train on any framework. All in plain Python, no ROS required.
How Positronic relates to LeRobot
LeRobot is excellent for learning ML robotics, prototyping policies, and running quick experiments on reference hardware.
Positronic builds on LeRobot's training infrastructure and adds what working systems need: hardware drivers for real robots, data ops that survive representation changes, a unified inference API across vendors, and iteration workflows. LeRobot gets you to a trained model. Positronic gets you to a robot that works.
Get involved
- 💬 Join Discord to ask questions, share setups, and request features.
- ⭐ Star on GitHub to follow our progress.
- ✉️ Email: [email protected]